Quantized Compressive Sensing And Its Application In IR-UWB Receiver

Impulse Radio Ultra-wideband （IR-UWB） possesses numerous advantages such aslarge system capacity, strong multi-path resolving capability. Especially because the lowpower and low cost characteristic, it gained a lot of attentions in recent years. However,the unprecedented bandwidth of IR-UWB demands for high-speed and high-precisionAnalog-to-Digital Converter （ADC）. It has become one of the major technicaldifficulties that hinder the practical application of IR-UWB technology.The recent research spot in applied mathematics fields—Compressive Sensing （CS）theory makes accurate signal reconstruction with sampling rate far below the Nyquist ratepossible, providing a natural solution for the ADC bottleneck issue. In the CS framework,through exploiting the sparsity of IR-UWB transmitted signal and propagation channel,low rate sampling of received IR-UWB signal can be achieved.This paper briefly introduces the CS theory and IR-UWB signal model firstly.Considering that transmitted IR-UWB pulse and the multi-path channel are essentiallysparse, the random demodulator architecture is adopted for IR-UWB signal compressedsampling. Undoubtedly, quantization noise is inevitable when the compressedmeasurements are transmitted from the analog front-end to the digital back-end.Obviously, it’s hard to achieve optimal signal reconstruction performance since thequantization noise characteristics were not exploited sufficiently in existent architecture.To improve the noise-robust capability of the reconstruction algorithm under CSframework, two excellent algorithms are selected from convex algorithm and greedyalgorithm in this paper. According to the characteristic of quantization noise,the signal reconstruction model is revised. Then a metric which can accuratelydistinguish the noise environment named QNR （quantization noise to thermal noise ratio）is defined. Based on the information of QNR, we propose a signal reconstructionmethod called Joint DS-SP, it can select algorithm adaptively between DS （DantzigSelector） and SP （Subspace Pursuit）. Compared with traditional algorithms, this methodcan achieve the best performance in different noise environments; moreover, thecomputational complexity of Joint DS-SP is between the DS method and SP method.This offers a new signal reconstruction way for the design of CS-based IR-UWBreceiver’s digital back-end.However, processing in the digital back-end cannot reduce the quantization noise atthe source. In order to improve the quantization noise-resistant performance in analogfront-end of IR-UWB system under CS framework, three modified quantizationmechanisms are proposed in this paper, they are: overload uniform quantization,non-uniform quantization and overload non-uniform quantization. The idea of these mechanisms are mainly originate from that all compressed measurements can carryinformation equally and their Gaussian distribution characteristics. The correspondingsolving models and algorithms are displayed in paper. Especially, we find an algorithmnamed PDIP （Primal Dual Interior Point） can solve the problem of l₁-l_∞. Thus, thereconstruction model of overload uniform quantization can be solved directly.With these bases, the optimization overload schemes are proposed in this paper.These overload schemes largely guaranteed the low complexity and excellentperformance of proposed mechanisms. Simulation results certified that all proposedmechanisms are obtaining obvious performance improvement compared to conventionaluniform quantization mechanism. Moreover, overload uniform quantization is suit formost practical application due to its low complexity. In proposed mechanisms, theoverload non-uniform quantization strategy has the optimal performance whichprovides a feasible solution for extreme high accurate application.